Ocean Sci., 10, 745–758, 2014
www.ocean-sci.net/10/745/2014/
doi:10.5194/os-10-745-2014
© Author(s) 2014. CC Attribution 3.0 License.
Characterisation and quantification of regional diurnal SST cycles
from SEVIRI
I. Karagali1 and J. L. Høyer2
1 DTU
Wind Energy, Technical University of Denmark, Risø Campus, Building 125, Roskilde, 4000, Denmark
for Ocean and Ice (COI), Danish Meteorological Institute (DMI), Lyngbyvej 100, 2100, Denmark
2 Centre
Correspondence to: I. Karagali ([email protected])
Received: 28 January 2014 – Published in Ocean Sci. Discuss.: 7 April 2014
Revised: 19 June 2014 – Accepted: 2 July 2014 – Published: 2 September 2014
Abstract. Hourly SST (sea surface temperature) fields from
the geostationary Spinning Enhanced Visible and Infrared
Imager (SEVIRI) offer a unique opportunity for the characterisation and quantification of the diurnal cycle of SST in the
Atlantic Ocean, the Mediterranean Sea and the northern European shelf seas. Six years of SST fields from SEVIRI are
validated against the Advanced Along-Track Scanning Radiometer (AATSR) Reprocessed for Climate (ARC) data set.
The overall SEVIRI–AATSR bias is −0.07 K, and the standard deviation is 0.51 K, based on more than 53 × 106 matchups. Identification of the diurnal signal requires an SST foundation temperature field representative of well-mixed conditions which typically occur at night-time or under moderate and strong winds. Such fields are generated from the
SEVIRI archive and are validated against pre-dawn SEVIRI
SSTs and night-time SSTs from drifting buoys. The different
methodologies tested for the foundation temperature fields
reveal variability introduced by averaging night-time SSTs
over many days compared to single-day, pre-dawn values.
Diurnal warming is most pronounced in the Mediterranean
and Baltic seas while weaker diurnal signals are found in
the tropics. Longer diurnal warming duration is identified
in the high latitudes compared to the tropics. The maximum
monthly mean diurnal signal can be up to 0.5 K in specific
regions.
1 Introduction
Diurnal warming of the sea surface temperature (SST) occurs under favourable conditions of solar heating and weak
winds, and it is most intense in the first few millimetres of the
water column; the part observable from microwave and infrared sensors. It has been observed in different areas of the
global ocean from the tropics and mid-latitudes (Merchant
et al., 2008; Price et al., 1987; Ward, 2006) to higher latitudes
(Eastwood et al., 2011; Karagali et al., 2012). Large diurnal
warming signals compared to drifting buoys have been reported in the inter-tropical Atlantic, whereas in other regions
of the “disc” (i.e. field of view) of the Spinning Enhanced
Visible and Infrared Imager (SEVIRI), the agreement between drifters and the satellite diurnal signal was found to be
around 0.5 K (Le Borgne et al., 2012b). Castro et al. (2013)
used SEVIRI SSTs and unpumped Argo floats to derive diurnal warming estimates, showing a good agreement between
the two measuring instruments.
Not accounting for the daily variability in SST may result
in biases of the total heat budget estimates (Webster et al.,
1996; Ward, 2006; Bellenger and Duvel, 2009; Bellenger
et al., 2010). Recently, Clayson and Bogdanoff (2013) estimated that the globally averaged error in the flux calculations
due to neglect of the SST diurnal signal is ∼ 4.5 W m−2 ,
but on an instantaneous basis, the total error can exceed
200 W m−2 in specific locations. Strong SST diurnal signals
can complicate the assimilation of SST fields in ocean and atmospheric models, the derivation of atmospheric correction
algorithms for satellite radiometers and the merging of satellite SST from different sensors (Donlon et al., 2007). In addition, SST gradients can cause biases in the scatterometerderived ocean winds and in the estimated net flux of CO2 ,
as the outflux of oceanic CO2 is positively correlated with
the increase of SST. Understanding the daily SST variability
is an ongoing scientific endeavour. Quantifying the diurnal
signals of SST at different regions, and resolving the vertical
Published by Copernicus Publications on behalf of the European Geosciences Union.
746
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
extent of the diurnal signal, is important in order to relate observations from different instruments and to remove trends
from climate records.
The aim of this study is to thoroughly characterise the
regional extent of diurnal SST signals over a range of latitudinal bands. The 6-year-long SEVIRI hourly SST fields
are used to perform a low, mid- and high latitude evaluation of the diurnal cycle and identify regional patterns. Identifying areas where common diurnal warming patterns occur is important to better understand the conditions under
which the diurnal cycle is formed. Envisat Advanced AlongTrack Scanning Radiometer (AATSR) SSTs hold a key role
for comparisons with the SEVIRI SSTs, especially in areas
where drifting buoys are not available. Focus is given on the
definition of the foundation temperature which must be representative of well-mixed conditions. As the diurnal signal
develops on the foundation temperature, an accurate estimation of the latter is required to avoid over- or underestimation
of the daily cycle’s amplitude.
The satellite and in situ data used in the study are described in Sect. 2, and the methodologies used for the collocation of data sets and the generation of night-time composite SST fields representative of foundation temperatures
are presented in Sect. 3. The results in this study are categorised as (i) SEVIRI–AATSR validation statistics found in
Sect. 4.1, (ii) sensitivity tests on the night-time composites
are presented in Sect. 4.2 and (iii) results on the regional diurnal warming signals can be found in Sect. 4.3. Finally, the
discussion regarding the key findings of the study is available
in Sect. 5 and conclusions are to be found in Sect. 6.
2
2.1
Data
SEVIRI
The Meteosat Second Generation (MSG) SEVIRI instrument is in geostationary orbit, centred at 0◦ N, 0◦ E, with
a nadir resolution of 3 km. Experimental hourly fields on
a 0.05◦ grid, from the Centre Météorologie Spatiale (CMS),
Météo France, were obtained for the period 2006–2011 in
order to analyse the regional diurnal warming in the SEVIRI disc. The selected domain extends from 73◦ W to 45◦ E
and 60◦ S to 60◦ N. The SEVIRI SSTs are corrected for
the cool skin bias by adding 0.2 K to the original skin retrievals and are therefore representative of sub-skin temperatures (O&SI SAF, 2006). SEVIRI SST retrievals are classified using a quality flag index of 0 (unprocessed), 1 (erroneous), 2 (bad), 3 (acceptable), 4 (good) and 5 (excellent).
In this study, only SEVIRI SSTs flagged with quality 3 or
higher were used.
Ocean Sci., 10, 745–758, 2014
2.2
AATSR
The AATSR, with a nadir resolution of 1 km, was on board
the polar orbiting ESA Envisat platform, operational from
2002 to 2012. Envisat had the local equatorial crossing
time (LECT) at 10:00, crossing from north to south during the descending orbit in daytime and from south to
north during the ascending orbit in night-time. The AATSR
Re-processing for Climate (ARC) data set v1.1 was obtained for the period January 2006–March 2010 and v1.1.1
from April 2010–December 2011, through the Natural Environment Research Council (NERC) Earth Observation
Data Centre (http://www.neodc.rl.ac.uk/browse/neodc/arc).
The selected file types are (i) daytime dual-view 2-channel
SST retrievals and (ii) night-time dual-view 3-channel SST
retrievals. Three different SST measurements are available
on a 0.1◦ grid, i.e. skin, sub-skin and depth, and an uncertainty flag is included in the file. In this study, we use the
AATSR skin measurements that have an uncertainty of 0.8 K
or lower.
2.3
Drifting buoy data
Temperature measurements from drifting surface buoys
that report observations on an hourly basis are obtained
from the Coriolis database (http://www.coriolis.eu.org/),
available for the entire Atlantic from 2006 to 2011. The
temperature sensor is placed at approximately 20 cm depth
(http://www.jcommops.org/dbcp/community/standards.html)
and the observations have an error of about 0.2 ◦ C (O’Carroll
et al., 2008).
3
3.1
Methods
SEVIRI–AATSR validation
The spatial and temporal matching of the SEVIRI–AATSR
SSTs is performed based on (i) a maximum 30 min difference
between local times, (ii) SEVIRI SST with quality flags ≥ 3
and AATSR SST with uncertainty ≤ 0.8 K and (iii) SEVIRI–
AATSR latitude and longitude difference ≤ 0.049◦ . To correct for the different reference depth of the AATSR and SEVIRI SSTs (skin vs. sub-skin), 0.2 K are subtracted from
each SEVIRI retrieval only for the validation part of this
study, and thus both AATSR and SEVIRI data sets are representative of SSTskin . The same holds when AATSR and SEVIRI are compared to drifter measurements. To examine the
sensitivity of the SEVIRI–AATSR statistics to a varying time
window, 5 min thresholds starting from 0 and up to 30 min
were defined. Statistics were computed for each threshold,
using only the SEVIRI–AATSR match-ups with time differences smaller or equal to that threshold.
www.ocean-sci.net/10/745/2014/
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
Table 1. Description of the parameters used for each one of the test
foundation fields (TFF) and the validation field (VF).
3.2
Name
Averaged days
(day±)
Night-time
window (LT)
Quality flag
TFF1
TFF2
TFF3
TFF4
TFF5
VF
7 (3)
1 (0)
1 (0)
1 (0)
1 (0)
1 (0)
00:00–04:00
00:00–04:00
00:00–04:00
00:00–04:00
00:00–06:00
last pre-dawn
3, 4, 5
1, 2, 3, 4, 5
3, 4, 5
5
3, 4, 5
5
SST foundation fields
In order to study the diurnal SST variability, a foundation
SST field (SSTfound ) representative of well-mixed conditions
in the upper oceanic surface layer is necessary. Test foundation fields (abbreviated to TFF) are composed from SEVIRI
night-time SSTs for the years 2006–2011 using a local nighttime window, different ranges of the MSG quality flags (abbreviated to QF) and averaging multiple or 1 day SSTs. See
Table 1 for a description of the different foundation fields.
Validation fields (VF) are composed daily from the last
pre-dawn value, which is considered representative of the
coldest SST, flagged with QF 5. The difference TFF–VF is
defined and the statistics are computed for each TFF–VF
combination. The “successful” TFF must combine low standard deviation and high data availability. For the latter reason, we do not use a combination of only QF 4, 5 for the
TFFs, as this would not mitigate the issue of potentially low
data availability (see number of available data N shown in
Table 2, rows 3–5). Such sensitivity tests aim at investigating
the impact of the different foundation fields with respect to
latitude. In addition, the Coriolis drifter data are used to generate drifter foundation fields (abbreviated to DTFF) similar
to the SEVIRI TFFs for the same period, for comparison.
4
4.1
Results
SEVIRI–AATSR validation
The verification of the match-up procedure and the sensitivity analysis regarding the time interval between SEVIRI
and AATSR retrievals (not shown) indicated a general decrease of the bias and the standard deviation with decreasing
time interval. The validation results are presented in terms
of the mean bias (µ), standard deviation (σ ) and correlation (r) between SEVIRI and AATSR in the top five rows
of Table 2. All match-ups satisfy the quality flag and uncertainty criteria. Statistics derived without using any uncertainty flag on the AATSR data produced higher data availability (by approximately 105 ) while the mean bias, standard
deviation and correlation coefficient remained the same. To
www.ocean-sci.net/10/745/2014/
747
Table 2. Statistics of SEVIRI, AATSR and Drifters. The mean bias
is defined as µ, the standard deviation as σ and the number of
match-ups, N. All match-ups are with AATSR uncertainties ≤0.8 K
included and the SEVIRI skin correction. Filtered match-ups are
based on µ within the interval µ ± 4 × σ . The upper five rows show
the statistics of SEVIRI–AATSR match-ups. The bottom three rows
include the statistics based on the availability of drifting buoys.
µ
σ
r
N
All
Filtered
SEVIRI QF 3–AATSR
SEVIRI QF 4–AATSR
SEVIRI QF 5–AATSR
−0.06
−0.07
−0.18
−0.06
0.03
0.56
0.51
0.55
0.46
0.45
0.996
0.997
0.997
0.998
0.997
53 393 988
53 127 984
22 195 034
7 262 765
23 670 185
SEVIRI–Drifters
AATSR–Drifters
SEVIRI–AATSR
−0.23
−0.16
−0.07
0.67
0.51
0.48
0.993
0.996
0.996
12 849
12 842
12 926
avoid contamination of spurious SST values, a filter is applied to exclude match-ups outside the interval µ ± 4 × σ
(rows 2–5).
Filtered match-ups indicate colder SEVIRI SST with µ =
−0.07 K, σ = 0.51 K and r = 0.997. A consistent decrease
of µ and σ is observed for increasing SEVIRI quality flags,
with only QF 5 data having a positive µ value. Using the
time and space collocation criteria for the existing match-up
database of SEVIRI–AATSR, drifting buoy measurements
are used as a cross-validation data set. The statistics between SEVIRI, AATSR and drifters are shown in the bottom three rows of Table 2. Note that the last row is different from rows 1 to 5 in that the SEVIRI–AATSR match-ups
have been calculated only where drifting buoy measurements
are available, for comparison. Moreover, the SEVIRI QF and
AATSR uncertainty values used for the comparisons with the
drifting buoys are the same as previously stated. SEVIRI–
Drifter biases are negative, indicating colder SEVIRI SSTs,
justified by the skin vs. 20 cm depth difference between the
two types of observations. Such is the case for the AATSR–
Drifter bias but the µ absolute value is lower. The SEVIRI–
Drifter σ is higher than the AATSR–Drifter σ and correlation
is marginally lower for SEVIRI–Drifter.
Figure 1 shows the µ and σ values of SEVIRI–AATSR filtered match-ups (row 2 of Table 2) binned in 25 km × 25 km
boxes. Negative biases exceeding −0.5 K are identified in the
tropics and the mid-latitudes of the North Atlantic. The standard deviation is generally less than 0.5 K for the largest part
of the domain except the western border of the disc, associated with the SEVIRI satellite zenith angle and the reduced
accuracy due to the large atmospheric path. There is an evident trend of increasing σ in the zonal direction from the
prime meridian which is not as obvious in the meridional
direction from the Equator, at least up to 60◦ N. Highest σ
values are found for the Labrador Sea, i.e the upper left corner of the domain, where SEVIRI SSTs and thus diurnal
Ocean Sci., 10, 745–758, 2014
748
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
Figure 1. SEVIRI–AATSR (a) mean bias µ, (b) standard deviation σ , (c) number of match-ups N, binned in 25 km × 25 km boxes.
warming estimates will be more uncertain. Most SEVIRI–
AATSR match-ups are in the European seas and around
southern Africa (Fig. 1c).
The ARC data set has well-documented and low biases of
the order of 0.3 K compared to in situ measurements (Embury et al., 2012). SEVIRI–AATSR biases for the filtered
and uncertainty-flagged data sets (see row 2 in Table 2)
agree with the biases and σ values reported in Le Borgne
et al. (2012a) before a bias adjustment was performed using
AATSR data. When only quality 5 SEVIRI data are used, µ
and σ are of the same order as in Le Borgne et al. (2012a)
after the bias adjustment. Daytime vs. night-time SEVIRI–
AATSR match-ups (not shown) indicate that for local times
extending 5 h around the AATSR equatorial crossing time,
i.e. for retrievals near the sub-satellite track, negative biases
mostly occur at night-time. From existing results in the literature (Embury et al., 2012; O’Carroll et al., 2008), AATSR
night-time retrievals have a lower error compared to daytime.
Our larger night-time biases indicate that SEVIRI is even
colder at night-time compared to AATSR. Le Borgne et al.
(2012b) also reported a higher night-time bias and standard
deviation of SEVIRI against drifting buoys and associated it
with the SEVIRI cloud mask performance, which is better at
daytime due to the use of the SEVIRI high resolution visible channel. This cold bias may arise at night-time, when
the visible channel is not available, because the cloud mask
may be identifying cloud covered pixels, which have lower
temperatures, as cloud free water pixels.
Negative SEVIRI–AATSR biases are mostly confined in
the tropics, and in Le Borgne et al. (2011) they were attributed to retrieval algorithm errors caused by the anomalous
atmospheric water vapour vertical profiles. The spatial distribution of the SEVIRI–AATSR biases shows strong positive
signals around upwelling regions and cold currents (Portugal, Canary, Benguela and Malvinas currents). In such areas the cold water temperatures can cause the atmosphere
to be cooler, and thus, with low atmospheric water vapour
content. The mechanism of the water vapour impact on SST
Ocean Sci., 10, 745–758, 2014
retrievals, as explained in Le Borgne et al. (2011), dictates
than since water vapour generally decreases the brightness
temperatures observed by infrared instruments, it results in
cool SST biases that are corrected for by the retrieval algorithms. In regions with low water vapour content, such as
cold current regions and upwelling areas, an over-correction
for water vapour may occur resulting in warmer SSTs, hence
the larger positive SEVIRI–AATSR biases. Especially off the
coast of Angola/Namibia and north-west Africa, the warm
SEVIRI bias may also be associated with an overestimation
in the SEVIRI dust index correction. Moreover, for the region of the cold Malvinas Current which is also at the edge
of the SEVIRI disc, accuracy is reduced. Similar spatial patterns of 10 day average mean SEVIRI–AATSR differences
are reported in Le Borgne et al. (2012a), averaged over boxes
of different dimensions (see their Fig. 3).
When SEVIRI and AATSR match-ups are further validated with drifter measurements, it is found that both sensors have a cold bias compared to drifters. Similar results,
though slightly lower than what we report here, were found
by Le Borgne et al. (2012a); SEVIRI before adjustment had
a µ of −0.08, σ = 0.57 K against AATSR and µ = −0.12,
σ = 0.56 K against drifters. Moreover, SEVIRI–Drifter biases reported in Le Borgne et al. (2012b), for 2011–2012,
were also marginally lower than what is reported in our study,
but the SEVIRI data in their study were subject to a correction using NWP (numerical weather prediction) fields for atmospheric water vapour. Our SEVIRI–Drifter bias and standard deviation are slightly larger than what is reported in
Castro et al. (2013) using pumped and unpumped Argo data.
This is associated with the higher accuracy and better depth
representation of the Argo floats compared to drifters, but
also with the 3-hourly SEVIRI product used in that study
which will tend to reduce noise in the measurements. The
impact of the SEVIRI quality flags in the statistics, found in
Castro et al. (2013), indicates a slight increase towards small
positive µ and slight improvement in the σ values for QF5
SST, consistent with our findings.
www.ocean-sci.net/10/745/2014/
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
Figure 2. Latitude dependent statistics of TFF–VF for the period
2006–2011.
The SEVIRI processing chain has recently been updated
to accommodate retrieval biases at some of the problematic
areas mentioned above. The new processing started in 2011
and up to now no re-processing of the SEVIRI archive is being performed, and thus this study uses the old data set. Some
of the well-documented biases found in this study are compensated for in the new data set (Le Borgne et al., 2012a).
However, since our focus is on reliable diurnal variability results from long and multi-year satellite data, the new data can
not be used at this point.
4.2
Validating test foundation SST fields
The averaged statistics for all the test foundation fields (TFF)
against the validation field (VF) and the drifter test foundation fields (DTFF), are shown in Table 3. The mean bias (µ)
for all the TFFs–VF is almost zero and standard deviation
(σ ) values are mostly less than 0.3 K, while very subtle differences between the TFF are identified. TFF–DTFF µ and
σ are slightly higher, showing warmer drifter values; this is
related to the cool skin effect, i.e. the increased heat loss to
the atmosphere, present in the SEVIRI SST but most likely
absent from the drifter measurements which are taken from
a reference depth of 20 cm. The lowest absolute mean bias
is found for TFF2 and TFF3 against the VF and for TFF4
against DTFF. Standard deviation of the TFF–VF is lowest
for TFF4 and TFF5, and for TFF1 when DTFFs are used.
The relatively high σ values (for TFF–DTFF) are associated
with the few night-time drifter data available for the composition of similar foundation fields to the SEVIRI ones, which
is also supported by the reduced σ found when a multi-day
composite (TFF1) is used.
Statistics of the TFF–VF for the entire period 2006–2011,
are presented as a function of the latitude in Fig. 2. For
all TFF–VF, µ is negative except for TFF4, which is only
www.ocean-sci.net/10/745/2014/
749
composed of quality 5 SST and shows positive µ mostly
around 0.05 K that only reach up to 0.1 K north of the Equator. This positive µ is associated with the quality flag used, as
also shown in row 5 of Table 2, but nonetheless that contribution is negligible. Ranging between 0.2 and 0.3 K, σ values
are the lowest for TFF4 and TFF5, especially in “noisy” areas
such as the higher latitudes of both hemispheres. Nonetheless, only slightly higher σ is found for TFF2 and TFF3. The
multi-day composite TFF1 (dark blue line) has the highest
absolute µ and σ , related to averaging over many days. The
data availability, defined as the percentage of available data
over the total amount of water grid cells in the domain, is
slightly reduced when TFF4 is used.
Comparing TFF2 and TFF3, the impact of the quality flags
is highlighted as including lower quality data in TFF2 causes
slightly higher σ compared to TFF3, especially between 30
and 5◦ S, and this is explained by the fact that TFF3 is composed only of quality 3 and higher. Comparing TFF3 and
TFF5, the impact of the night-time window is revealed as
TFF5 has a marginally smaller absolute bias in some latitude
bands compared to TFF3 while σ is lower by 0.05 K. The
slightly improved statistics of TFF5 are related to the longer
night-time window (00:00–06:00 LT) that includes the predawn SST value of which VF is composed.
The spatial distribution of the µ, its σ and number of
match-ups for the period 2006–2011 is shown in Fig. 3 for
TFF3–VF. The data availability is defined as the number of
available TFF–VF match-ups at each grid cell and the number can not exceed a total of 2190, i.e. one match-up per day
for 6 years. A threshold was applied to avoid spurious SSTs,
hence grid cells with less than 10 TFF–VF match-ups are excluded. µ ranges between −0.1 and 0.1 K while σ values are
around 0.3 K for most parts of the SEVIRI disc. Higher µ
and σ are observed in areas with currents, associated with
high SST variability, and at the western boundary of the disc.
In summary, the validation of the various night-time SST
composites against single-day, pre-dawn SSTs assumed to
represent the coldest temperature during a day shows an absolute µ of the order of ±0.1 K and σ values not exceeding 0.3 K. Thus, the night-time fields can accurately represent cold, night-time foundation temperatures. Using only
quality 5 data results in slightly warmer foundation temperatures. This is in accordance with findings from the SEVIRI–
AATSR validation, which showed that quality 3 or 4 SEVIRI
data are colder than quality 5, and is associated with the cloud
masking as lower quality data have higher chances of cloud
contamination, i.e. lower pixel SST.
Validation with drifting buoy night-time composites has
shown that there is a consistent pattern of colder SEVIRI
foundation fields, reaching a maximum of −0.17 K due to
the difference reference depth. This bias is minimum for
TFF4, which consists of only quality 5 SEVIRI SST. Castro
et al. (2013) stated that underestimating the SEVIRI foundation fields compared to Argo fields showed a smaller impact on the diurnal warming estimates than overestimating
Ocean Sci., 10, 745–758, 2014
750
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
Table 3. Averaged statistics of TFF–VF and TFF–DTFF, for the period 2006–2011. The variability from the N of TFF1 is defined
−NTFF1
as NTFFX
× 100.
N
TFF1
µ [K]
TFF1
TFF2
TFF3
TFF4
TFF5
σ [K]
N
N (×109 )
Variability from
VF
DTFF
VF
DTFF
VF (×109 )
DTFF(×106 )
TFF
NTFF1 (%)
−0.05
−0.01
−0.01
0.06
−0.02
−0.12
−0.16
−0.17
0.01
−0.17
0.35
0.30
0.29
0.24
0.25
0.64
0.73
0.65
0.67
0.67
0.22
0.18
0.19
0.14
0.21
3.77
0.15
0.15
0.05
0.24
3.54
1.09
1.06
0.40
1.26
0%
−69.31 %
−69.95 %
− 88.51 %
−64.32 %
Figure 3. Mean bias µ (a), standard deviation σ (b) and number of match-ups N (c), for the TFF3–VF.
the foundation fields. From results not shown here, it was
found that using different foundation fields to obtain diurnal
warming estimates does not significantly impact the identified warming, in terms of its amount (in hours), its seasonal
and spatial distribution and the peak warming of cases exceeding 2 K. Given the bias statistics, and accounting for low
µ and σ combined with a sufficient data availability, TFF3
has been chosen as the foundation field to be used for the
regional diurnal warming analysis.
In Le Borgne et al. (2012b), the bias between SEVIRI and
drifter foundation fields was −0.12 K and the standard deviation 0.39 K, but that study used 1 year of upgraded SEVIRI
data with the NWP correction and defined foundation fields
from 00:00 LST to sunrise. Foundation fields composed of
night-time SEVIRI SSTs from 22:00 to 07:00 local solar time
(LST) in Castro et al. (2013) had smaller biases and standard
deviations compared to Argo foundation fields than what is
presented in this study, which is attributed to the more accurate Argo measurements compared to the drifters and the
3-hourly satellite SST product used. Moreover, they reported
slightly smaller biases when only quality 5 data were used
with reduced data availability, in accordance with our findings. Finally, for diurnal warming estimates, biases relative to
buoys are not a major concern since diurnal warming is estimated from SEVIRI daytime observations vs. SEVIRI foundation fields.
Ocean Sci., 10, 745–758, 2014
4.3
Diurnal warming
The difference, δSST, between a given daytime hourly SST
value and the corresponding foundation temperature of the
previous night is defined as diurnal warming or anomaly.
Given the biases and standard deviations found from the validation of the SEVIRI TFFs against Drifter TFFs, a threshold for warming of at least 1 K or more is used. The spatial extent of anomalies exceeding 1 K, in hours, is presented
in Fig. 4a for the period 2006–2011. Areas prone to diurnal warming are the European seas, the Red Sea, the central North and South Atlantic, and the areas of the Malvinas/Brazil currents (SW Atlantic), the Benguela Current
offshore from Namibia and Angola (SE Atlantic) and the
Mozambique Channel (Indian Ocean). As Fig. 4b shows,
most of the quality 5 SST retrievals are also located in these
areas. Features such as the mistral wind pattern in the eastern
part of the Gulf of Lion and the Etesian (Meltemi) winds
in the Aegean Sea, appear as areas of few diurnal warming cases, consistent with what is expected in these areas of
strong summer winds.
The spatial extent of cases for which δSST ≤ −1 K is
shown in Fig. 4c and is considered a measure of random
noise and oceanic variability not related to the daily warming
cycle, as it represents the occurrences when daytime SST is
lower than the night-time foundation field. The occurrence of
www.ocean-sci.net/10/745/2014/
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
751
Figure 4. Spatial extent of (a) hours of δSST ≥ 1 K for 2006–2011 where white areas indicate zero occurrences, (b) quality 5 data from 08:00
to 22:00 LT and for 2006–2011, (c) number of cases with δSST ≤ −1 K at any time between 08:00 and 22:00 LT for 2006–2011 where zero
occurrences are shown as white areas and the domains where regional diurnal warming is studied are plotted as boxes and (d) concurrences
of ECMWF wind speed ≤ 6 m s−1 and surface net solar flux ≥ 400 W m2 .
such cases is extremely rare in the entire SEVIRI disc, with
the exception of the Malvinas Current region in the western
South Atlantic and the eastern Gulf of Lion in the Mediterranean Sea. Areas with some few occurrences include the
Alboran Sea (east of the Gibraltar Straight), the Ionian and
Levantine seas and the northern Red Sea. In these regions,
diurnal warming estimates will have a slightly higher uncertainty compared to other areas but generally negative δSST
cases do not exceed 40 occurrences. Strong ocean features
such as the Gulf Stream and the Agulhas Current do not appear as areas with negative anomalies, and this is probably
related to the fact that SST variability in these areas occurs
at longer time-scales.
Using 3-hourly model outputs from the European Centre for Medium-Range Weather Forecasts (ECMWF), the
number of concurrent wind speeds ≤ 6 m s−1 and insolation ≥ 400 W m−2 for the period 2009–2011 was estimated,
shown in Fig. 4d. The spatial patterns of high occurrences
generally match the diurnal warming patterns. An exception is the area offshore from west Africa, between 15◦ S
www.ocean-sci.net/10/745/2014/
and 15◦ N, where despite the relatively high concurrence
of weak winds and strong solar heating, quality 5 SST and
δSST ≥ 1 K are mostly absent. This absence of quality SSTs
may be related to the cloud masking scheme. The mismatch between ECMWF low wind/high insolation occurrences and relatively strong diurnal warming signals off the
Namibia/Angola coastline is attributed to advection of warm
air masses from the nearby continent which can warm up the
top layer of the ocean surface. Evidence of this circulation
pattern is demonstrated in Richter and Mechoso (2004).
To obtain an estimate of the impact diurnal warming
can have on monthly mean SST estimates, the difference
1SSTday is defined as the mean SST during the “day time”
(i.e. all SSTs not included in the foundation field) minus
the SSTfound of that day. This provides a daily mean diurnal
warming estimate which is averaged monthly; thus, 12 mean
monthly estimates, 1SSTmonth , are obtained. For each grid
cell, the maximum 1SSTmonth is shown in Fig. 5a, excluding grid cells with less than 10 1SSTday values. The vertical
stripes visible in the centre of the disc are an artefact of the
Ocean Sci., 10, 745–758, 2014
752
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
local timing, as grid cells on the east side of a time zone are
at a later time and thus potentially have higher temperatures
compared to the ones on the west side of the same time zone.
The figure shows that diurnal warming can have a significant impact on monthly SST estimates, if not accounted for.
Average diurnal warming exceeding 0.4 K is identified in extended parts of the SEVIRI disc, not only in enclosed basins
like the Mediterranean, Black, North and Baltic seas but also
in open ocean waters. The σ of the mean diurnal warming
signal is estimated separately in each grid cell from all the
1SSTday of the month with the maximum signal. Figure 5b
shows values in the range of 0.6–0.8 K in areas where peak
mean diurnal warming is found.
Based on the frequency of δSST ≥ 1 K, eight sub-domains
were identified in the SEVIRI disc and are shown in Fig. 4c.
They are classified as mid-latitude domains (ML1, ML2),
sub-tropics (ST1, ST2, ST3), tropics (T1) and European seas
(ES1, ES2). Diurnal warming was studied in these regions
in terms of the monthly distribution exceeding some thresholds, the mean local time of occurrence, the distribution
of δSST ≥ 2 K, the duration of warming and the average
monthly cycles.
Only for the cases with δSST ≥ 2 K, Table 4 shows the
75th and 95th percentile δSST values and their duration. The
75th percentiles vary little between domains and range from
2.6 to 2.9 K but the 95 % have a wider distribution, from
3.2 K in the tropics to 4.0 K in the higher mid-latitudes of
the Northern Hemisphere. In particular, domains ML1 and
ES2 show almost the same percentiles of high warming cases
even though Fig. 4a suggests a dramatic decrease in identified warming on ML1 compared to ES2. In almost all domains, 75 % of the cases exceeding 2 K do not last more than
3 h, with the exception of the tropics and sub-tropics where
the 75th percentile is lowered to 2 h and the North/Baltic
seas, where it increases to 4 h. Similar is the case for the 95th
percentiles, ranging from 4 h close to the Equator to 6 h in the
mid-latitudes but reaching up to 9 h in the North/Baltic seas.
4.3.1
Mid-latitudes
Monthly distributions of δSST ≥ 1, 2 and 3 K for domains
ML1 and ML2 are shown in Fig. 6a and b. Most δSST cases
occur during the Northern and Southern Hemisphere summers. In domain ML1 the distribution has a main peak in
June and July. Domain ML2 has a less varied signal, as ∼ 12–
14 % of δSST ≥ 1 occur from November to February.
The distributions for the local time of δSST ≥ 2 K occurrences, in Fig. 7a and b, have a peak at 15:00 LST. Both areas
are geographically located in the upper mid-latitudes of the
North and South Atlantic, where during summer the long period of sunlight drives early warming, more pronounced in
domain ML1 compared to ML2.
Averaged hourly diurnal cycles are calculated from grid
cells where δSST exceeds 0.5 K at least once during the
day. In each domain, the averaged diurnal cycles of the
Ocean Sci., 10, 745–758, 2014
4 months with the highest number of recorded δSST ≥ 2 K
are shown in Fig. 8. The averaged daily cycle in domain
ML1 (Fig. 8a) shows a peak of ∼ 1.2 K. Minimum values
are 0.1 K higher than the foundation temperature between
midnight and 02:00 LST, with warming already starting after that point. A smooth warming peaking between 14:00–
17:00 LST is observed but cooling occurs faster. Domain
ML2 (Fig. 8b) has a smoother cycle with a lower peak for
all selected months. The morning minimum between 02:00
and 03:00 LST almost reaches foundation temperature values while peak warming occurs between 14:00–16:00 LST.
Residual warming of 0.4–0.6 K is observed in both domains,
but particularly for ML1 results are ambiguous as the amount
of δSST ≥ 1 K is limited.
4.3.2
Sub-tropics
The monthly distribution of δSST exceeding different thresholds indicates that in the North Atlantic domain ST1
(Fig. 6c), a clear seasonal signal exists with a peak of δSST
in June, and almost no cases exceeding 2 K for the period
October–February. South Atlantic domain ST2 (Fig. 6d) has
in total a high number of δSST ≥ 1 between November and
February, reaching up to 17 % in December and January.
Cases with δSST ≥ 1 decrease significantly between May
and August while no more than 2 % of the cases exceed 2 K
from March to October. Similar is the pattern for domain ST3
(Fig. 6e) but with lower percentages and a secondary peak in
October.
Warming exceeding 2 K peaks in all three domains at
14:00 LST (Fig. 7c–e), with almost no differences at the start
of the cycle. In all three domains warming may last up to
23:00 LST, slightly more pronounced in South Atlantic ST2,
where more warming occurs compared to the other domains
(see Fig. 4a).
This is confirmed by the monthly average diurnal cycles
shown in Fig. 8. Domain ST1, in Fig. 8c, shows a very stable cycle throughout the months, with a morning minimum
around 04:00 LST and a peak of 1 K occurring at 14:00 LST.
Cooling occurs smoothly and a residual warm layer of 0.4 K
is observed as the temperature at night-time does not return to
the pre-warming value. Domain ST2 (Fig. 8d) shows a nighttime cooling that reaches a minimum at 03:00 LST and remains at minimum until 06:00 LST, being slightly colder
than the foundation temperature. The peak warming ranges
between 1 and 1.2 K, depending on the month, occurring between 14:00–15:00 LST. ST3 (Fig 8e) has similarities with
both ST1 and ST2, showing a long morning minimum period
around the foundation temperature value and rapid cooling.
It is subject to a higher peak δSST that varies depending on
the month. In all domains, a residual warm layer of 0.3–0.4 K
is observed.
www.ocean-sci.net/10/745/2014/
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
753
Figure 5. Spatial extent of (a) 1SST, defined as the maximum of the monthly averaged 1SSTday , i.e. the mean SSTday − SSTfound , where
white areas indicate less than 10 1SSTday values, (b) σ of the 1SSTday during the month when 1SST occurs.
Table 4. Percentiles of δSST ≥ 2 K and its duration (in hours), for
the different domains.
ML1
ML2
ST1
ST2
ST3
T1
ES1
ES2
4.3.3
δSST ≥ 2 K
Duration (hours)
75 %
95 %
75 %
95 %
2.9
2.7
2.7
2.6
2.8
2.6
2.8
2.8
4.0
3.7
3.3
3.3
3.7
3.2
3.7
3.9
3
3
3
2
3
2
3
4
7
7
5
5
4
5
5
9
Tropics
The tropical domain (T1) is the region with an almost nonexistent seasonal variability, a constant 2–3 % of δSST ≥ 1 K
during the year and only very few cases of δSST ≥ 2 K
(Fig. 6f). The few δSST ≥ 2 K cases start at ∼ 08:00 LST,
peak at 14:00 LST (Fig. 7f) and last until late at night. The averaged monthly cycle, shown in Fig. 8f, shows a very smooth
warming from the morning minimum at 01:00–02:00 LST to
a peak of 0.6 K at 14:00 LST. Some changes are observed
depending on the month; warming starts earlier in April and
May, the morning minimum is longer and more pronounced
in February and March and the peak is slightly higher and
occurs later compared to the April–May case. Cooling generally occurs slowly and the late night temperature decreases
to almost pre-warming levels.
4.3.4
European seas
A clear seasonal trend is revealed in the Mediterranean/Black
Sea domain (Fig. 6g), with a peak in δSST ≥ 1 K during
www.ocean-sci.net/10/745/2014/
May approaching 10 % of the quality 5 SSTs. The frequency
of δSST ≥ 2 K is almost constant from May to July. The
North/Baltic seas domain has a more prominent seasonal signal for all thresholds, with a distinct peak in June (Fig. 6h).
Cases of δSST ≥ 2 K reach up to 4 % of the total quality 5
SSTs, the highest percentage of all domains.
The local time of occurrence shows a shift between the two
domains, as the prime time for warming in ES1 is 13:00 LST
(Fig. 7g), with practically no occurrences before 09:00. Domain ES2 has a wider distribution in the time of occurrence
(Fig. 7h), with a peak at 16:00 LST and cases as early as
08:00 LST.
The monthly average diurnal cycle in domain ES1, shown
in Fig. 8g, has a prominent night-time cooling resulting in
a negative minimum at 05:00 LST, showing that early morning SST can be colder than the SSTfound . Warming is fast
and the peak value changes depending on the month, reaching a maximum up to 1.3 K in May. A residual warm layer
of 0.4–0.5 K is maintained independent of the month. At
higher latitudes, domain ES2 also has a morning cooling period that results in negative anomalies only during June and
July, but lasts for a shorter period, peaks at 04:00 LST and
is completely suppressed in May and August (Fig. 8h). Peak
warming reaches 1.4 K in June but it varies depending on
the month and in August, the peak value is 0.9 K. A residual
warm layer ranging from 0.5–0.6 K in May and August, to
0.7 K in June and July is present.
5
Discussion and summary
The main focus of this study is to characterise regional diurnal warming on the SEVIRI disc. In order to obtain accurate
and trustworthy diurnal estimates, SEVIRI SSTs were validated against AATSR SST retrievals, highlighting regions
where SEVIRI retrievals are of high quality but also regions
Ocean Sci., 10, 745–758, 2014
16
Karagali & Høyer: SEVIRI Regional Diurnal Warming
754
10
5
0.1
10
5
0.08
0.06
0.04
0.02
0
2
3
4
5
6
7
Month
8
9
10
11
0
12
20
2
3
4
5
6
7
Month
8
9
10
11
12
>=1 K (19159885)
>=2 K (1658006)
>=3 K (168277)
ST1
Normalized Count [%]
15
10
5
1
2
3
4
5
(c)
6
7
Month
8
9
10
11
15
10
1
2
3
4
5
6
7
Month
8
9
11
12
5
0.1
0.2
15
5
3
4
5
6
7
Month
8
9
10
11
>=1 K (13041705)
>=2 K (405510)
>=3 K (35344)
10
12
1
3
4
5
6
7
Month
8
9
10
11
ST3, N.=1615295
5
0.1
3
4
5
6
7
Month
8
9
10
11
Normalized Count
15
10
(h)
0.1
0.05
5
10
15
20
Mean Local Time of δSST >= 2[K]
0
5
(f)
10
15
20
Mean Local Time of δSST >= 2[K]
0.12
0.14
ES2, N.=2052089
0.12
0.1
0.08
0.06
0.1
0.08
0.06
0.04
0.04
0.02
0.02
5
0
12
0.15
0.16
ES2
0
(g)
2
10
15
20
Mean Local Time of δSST >= 2[K]
0.2 T1, N.=405510
0.18 ES1, N.=8942132
>=1 K (11543511)
>=2 K (2052089)
>=3 K (409317)
>=1 K (76248159)
>=2 K (8942132)
>=3 K (1561801)
1
5
12
20
Normalized Count [%]
Normalized Count [%]
2
(f)
10
0
0
0.05
ES1
(g)
0.1
5
0
20
15
0.15
(d)
Normalized Count
2
10
15
20
Mean Local Time of δSST >= 2[K]
0.15
0
1
10
15
20
Mean Local Time of δSST >= 2[K]
0.05
(e)
(e)
5
0.2 ST2, N.=1658006
0.15
(c)
Normalized Count
>=1 K (11889919)
>=2 K (1615295)
>=3 K (316736)
10
0
0
T1
Normalized Count [%]
Normalized Count [%]
10
20
ST3
15
0.04
(b)
0.05
(d)
20
0.06
5
0
12
10
15
20
Mean Local Time of δSST >= 2[K]
0.2 ST1, N.=1280834
ST2
0
0
5
(a)
20
>=1 K (12300240)
>=2 K (1280834)
>=3 K (159075)
Normalized Count [%]
1
(b)
0.08
0.02
Normalized Count
1
Normalized Count
0
(a)
0.1 ML2, N.=2214690
ML1, N.=431956
15
Normalized Count
15
ML2
Normalized Count
>=1 K (16278193)
>=2 K (2214690)
>=3 K (347303)
ML1
Normalized Count [%]
Normalized Count [%]
0.12
20
>=1 K (2797304)
>=2 K (431956)
>=3 K (90505)
Normalized Count
20
& Høyer: SEVIRI Regional Diurnal Warming
15
I. Karagali
and J. L. Høyer: SEVIRI Karagali
regional
diurnal warming
1
2
3
4
5
6
7
Month
8
9
10
5
10
15
20
Mean Local Time of δSST >= 2[K]
0
(h)
5
10
15
20
Mean Local Time of δSST >= 2[K]
Fig. 7. Distribution of mean local time for δSST ≥ 2 K, in different domains of the SEVIRI disc: (a) ML1, (b) ML2, (c) ST1, (d) ST2, (e)
12
ST3, (f) T1, (g) ES1 and (h) ES2.
11
Figure 7. Distribution of mean local time for δSST ≥ 2 K, in difdomains of the SEVIRI disc: (a) ML1, (b) ML2, (c) ST1,
(d) ST2, (e) ST3, (f) T1, (g) ES1 and (h) ES2.
Fig. 6. Monthly distribution of hourly δSST exceeding 1, 2 and 3 K from 2006–2011, normalised over the total number of quality 5ferent
SST per
month in domains:
(a) ML1, (b)
(c) ST1, distribution
(d) ST2, (e) ST3, (f)
(g) ES1 and
(h) ES2.
Figure
6. ML2,
Monthly
ofT1,hourly
δSST
exceeding 1, 2 and
3 K from 2006–2011, normalised over the total number of quality 5
SST per month in domains: (a) ML1, (b) ML2, (c) ST1, (d) ST2,
(e) ST3, (f) T1, (g) ES1 and (h) ES2.
where diurnal signals will be noisy and thus uncertain. In
particular, a zonal increase of the SEVIRI–AATSR standard
deviation with distance from the prime meridian is observed,
associated with the SEVIRI reduced retrieval accuracy due
to the larger atmospheric path, but this is not as obvious in
the meridional direction as the distance from the Equator increases. Prior to the estimation of the daily signal, a detailed
analysis on the impact of the foundation field was performed.
It has been shown that the foundation fields generated from
night-time SEVIRI SSTs are representative of well-mixed
conditions and thus are appropriately used as the basis upon
which the diurnal cycle is developed.
Until now, no systematic investigation has been performed
on the derivation of representative foundation fields. Studies mentioned previously generally utilise night-time SSTs
of varying quality flags and time-windows, and compare the
Ocean Sci., 10, 745–758, 2014
composites with in situ measurements. In this study, a thorough sensitivity test was performed, by generating various
foundation fields and validating them with pre-dawn satellite SSTs and drifting buoy measurements. It was shown
that marginally better statistics are obtained when using only
quality 5 data but availability is reduced. In addition, averaging over many days results in more gap-free foundation fields
but it also introduces more variability, as shown from the increasing σ values. Finally, longer night-time windows do not
necessarily guarantee more data availability while statistics
remain the same, but during summer at high-latitudes some
daytime SSTs will be included due to the long daytime period.
Diurnal warming was identified in large areas of the SEVIRI disc, particularly in enclosed seas but also the open
ocean while a somewhat suppressed cycle was found in the
intertropics. This latter finding may be related to the physical conditions prevailing outside of the Inter-Tropical Convergence Zone (ITCZ), characterised by high and relatively
www.ocean-sci.net/10/745/2014/
Karagali
SEVIRI
Regional
Diurnal
Warming
I. Karagali
and J.&L.Høyer:
Høyer:
SEVIRI
regional
diurnal
warming
1.6
1.2
1.4
1.2
δSST [K]
δSST [K]
ML2
1
0.8
0.6
0.8
0.6
0.4
0.4
0.2
0.2
0
0
0
2
4
6
8
(a)
10 12 14
Local Time
16
18
20
22
24
0
2
4
6
8
(b)
1.6
10 12 14
Local Time
16
18
20
22
24
16
18
20
22
24
16
18
20
22
24
16
18
20
22
24
1.6
May
Jun
Jul
Aug
1.4
1.2
Jan
Feb
Nov
Dec
ST1
1.4
1.2
ST2
1
δSST [K]
1
δSST [K]
Jan
Feb
Nov
Dec
ML1
1
0.8
0.6
0.8
0.6
0.4
0.4
0.2
0.2
0
0
0
2
4
6
8
(c)
10 12 14
Local Time
16
18
20
22
24
0
Jan
Feb
Oct
Dec
1.2
6
8
Feb
Mar
Apr
May
ST3
1.4
1.2
10 12 14
Local Time
T1
1
δSST [K]
1
0.8
0.6
0.8
0.6
0.4
0.4
0.2
0.2
0
0
2
4
6
8
(e)
10 12 14
Local Time
16
18
20
22
24
0
2
4
6
8
(f)
1.6
10 12 14
Local Time
1.6
Apr
May
Jun
Jul
1.4
1.2
May
Jun
Jul
Aug
ES1
1.4
1.2
ES2
1
δSST [K]
1
0.8
0.6
0.8
0.6
0.4
0.4
0.2
0.2
0
0
4
1.6
1.4
0
2
(d)
1.6
δSST [K]
755
1.6
May
Jun
Jul
Aug
1.4
δSST [K]
17
0
2
(g)
4
6
8
10 12 14
Local Time
16
18
20
22
24
0
(h)
2
4
6
8
10 12 14
Local Time
Fig. 8. Monthly-averaged daily cycles in different domains of the SEVIRI disc: (a) ML1, (b) ML2, (c) ST1, (d) ST2, (e) ST3, (f) T1, (g)
Figure 8. Monthly
averaged
daily
cycles
different
the SEVIRI
disc:
(a)are
ML1,
(b)TheML2,
(d) ST2,
(e) ST3, (f) T1,
ES1 and (h)
ES2. In each
domain,
thein
4 months
withdomains
the highestofnumber
of recorded
δSST
shown.
daily (c)
cycleST1,
is produced
by hourly
(g) ES1 andaveraging
(h) ES2.
each
the≥40.5
months
with
the
highest
number
recorded
are
shown.
The daily
is produced by
gridIncells
thatdomain,
show δSST
K at least
once
during
the day.
A zeroof
value
indicatesδSST
that the
monthly
averaged
hourlycycle
SST value
is equalgrid
to thecells
foundation
temperature
value.
the end
of day.
the cycle
doesvalue
not approach
the pre-warming
value,
a residual
hourly averaging
that show
δSST ≥
0.5 IfKthe
at temperature
least once at
during
the
A zero
indicates
that the hourly
SST
value is equal to
warmtemperature
layer exists. value. If the temperature at the end of the cycle does not approach the pre-warming value, a residual warm layer
the foundation
exists.
stable water temperature, constant trade winds, convection
and high precipitation rates. Nonetheless, a large area off
the western Sahara showed relatively strong diurnal warming signals. Since the SEVIRI–AATSR differences are very
low in this area, SEVIRI is considered accurate there. Thus,
it is postulated that the very high water turbidity, verified by
www.ocean-sci.net/10/745/2014/
composite maps from the GlobColour project (Morel et al.,
2007), reduces the incoming solar radiation’s penetration
depth in the water column thus strengthening the top layer
warming.
Ocean Sci., 10, 745–758, 2014
756
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
The characteristics of the diurnal cycle differ depending on the region, in terms of the seasonal distribution, the
amount of warming exceeding certain thresholds, the peak
local time of occurrence and the average monthly diurnal
cycle. Consistent findings include the occurrence during the
spring/summer months in each hemisphere, a dramatic difference between identified δSST exceeding 1 and 2 K, a peak
occurring a few hours after local noon with the exact timing
shifting towards later hours with increasing latitude, an early
morning cool period and a residual warm layer at night-time.
Results from Le Borgne et al. (2012b) using 1 year of the
upgraded hourly SEVIRI SST over the entire Atlantic indicated a diurnal cycle showing a morning cooling period with
a minimum around 06:00 LST, a warming period peaking at
15:00 LST and a residual night-time warm layer smaller than
0.5 K. Our results generally agree, but being area specific,
they show that the morning minimum is more pronounced
in the mid-latitudes compared to the sub-tropics and that the
cycle may reach a peak of 1.4 K in the North/Baltic seas,
compared to a moderate 0.6 K in the tropics.
Peak monthly diurnal warming of 0.3 K was identified
in extended areas of the Atlantic Ocean and the European
seas, but estimates up to 0.5 K were also found. Peak diurnal warming over a 3-year period, as shown in Castro et al.
(2013), had similar spatial distributions as in our study but
higher values were recorded. Kennedy et al. (2007) calculated a climatology of diurnal warming estimates from drifting buoy data expanding from 1990 to 2004, where annual
average warming patterns and spatial distributions were similar to our findings. However, the dense coverage of the
SEVIRI observations provides more detailed estimates of
the spatial and temporal variability of the diurnal warming
events.
Prytherch et al. (2013) in five mooring sites, mostly out of
our domain, observed mean diurnal warming of the same order as reported here. Specifically for a station in the Red Sea
and for 2 years of observed temperatures, they stated a mean
maximum warming of 0.43 K while findings from our Fig. 5
indicate a 1SST of 0.17 K which reaches up to 0.5 K when
σ is added. In the Mediterranean Sea, Marullo et al. (2014)
computed mean diurnal cycles for 3 months in the summer of
2011 showing a SEVIRI peak warming of ∼ 1.2 K occurring
at 15:00–16:00 LST, very much in agreement with our study
despite the differences arising from the newly processed SEVIRI data set used, their much shorter period of study and
their averaging over all the summer months to produce the
daily cycle.
Such findings provide good arguments regarding the validity of our results. Nonetheless, SEVIRI SSTs can be biased cold in areas with anomalous atmospheric profiles. Subpixel clouds generated during the afternoon will have a similar cold effect, of the order of 0.2 K, on the diurnal warming estimates particularly for large signals, as demonstrated
in Le Borgne et al. (2012b). Increased afternoon convection
when diurnal layers are created has also been reported by
Ocean Sci., 10, 745–758, 2014
Bellenger et al. (2010). Using the newly processed SEVIRI
archive of SST including the NWP model correction for water vapour (Le Borgne et al., 2011) may provide some different results regarding the derived diurnal warming estimates
from the updated SEVIRI data. Nonetheless, it is justified to
assume that changes in the general spatial distribution, the
timing and magnitude of the signal are expected to be minor.
Therefore, results from this study provide a first and reliable
description of the diurnal cycle characteristics over several
domains and can be useful for satellite SST users that wish
to account for or model the diurnal signals in their areas of
interest.
6
Conclusions
A comprehensive description of diurnal warming estimates
in the Atlantic Ocean and the European Seas is presented,
based on 6 years of hourly SST observations. In addition,
an extensive sensitivity analysis on the definition of foundation temperature fields is performed to evaluate the impact
on the diurnal variability estimates. Finally, SEVIRI data are
compared with the AATSR re-processed for climate archive
(ARC) to establish a quality control on the geostationary sensor which, particularly at large angles, has lower accuracy
and resolution. The mean SEVIRI–AATSR bias is −0.07 K
with a standard deviation of 0.51 K. In the mid-latitudes of
both hemispheres, SEVIRI–AATSR biases are almost zero
and the standard deviation shows minimum values; in the
tropics and high latitudes the bias becomes negative.
Prior to the estimation of diurnal signals, test foundation SST fields are composed from SEVIRI night-time SSTs
and are validated against SEVIRI pre-dawn SSTs and nighttime composites from drifting buoys. The validation of SEVIRI foundation fields against pre-dawn SSTs shows almost
zero bias and standard deviations of 0.3 K indicating a good
description of night-time, mixed conditions. Validated with
drifter composites, it is found that SEVIRI foundation fields
are, on average, slightly colder. Generally higher σ values
are observed in areas with currents associated with high SST
variability and at the western boundary of the SEVIRI disc.
However, for diurnal warming estimates, biases relative to
buoys are not a major concern as the diurnal warming results
are derived from SEVIRI daytime observations vs. SEVIRI
foundation fields.
An impact of the SEVIRI quality flags is identified, where
quality 5 data are warmer and show better statistics with
drifter composites. Thus, the SEVIRI–Drifter differences are
partly associated with the possible partial cloud coverage of
SEVIRI pixels for qualities of 4 and lower. Another bias contribution arises from the reference depth of drifting buoys
(∼ 20 cm) and SEVIRI SSTs (sub-skin estimated as skin
+0.2 K). Cool skin effects are largest for weak winds, which
is also the case for diurnal warming events. The impact of
the large skin effect would be to reduce diurnal warming due
www.ocean-sci.net/10/745/2014/
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
to heat loss to the atmosphere. Sub-pixel clouds generated
during the afternoon will have a similar cold effect. These
factors suggest that our diurnal warming estimates may be
conservative regarding the peak value and offset.
Diurnal warming estimates extend over the entire SEVIRI disc and generally match well with the spatial pattern of the regions with low wind and high insolation from
ECMWF fields. The impact of diurnal variability on monthly
mean SST estimates is assessed for the full study area. The
maximum monthly mean influence from diurnal variability
reaches up to 0.5 K in extended areas of the SEVIRI disc,
mostly confined close to continents and in enclosed seas.
Consistent patterns between the diurnal warming estimates
include the seasonality of the signal being constrained by
the spring/summer season, the morning cooling and residual warm layer. But regional differences in the warming patterns are also identified, particularly regarding the warming
thresholds, peak values and timing. It is outside the scope of
the present paper to investigate in detail the mechanisms that
lead to such regional differences in the diurnal cycle. Water turbidity may hold a role in promoting the development
of a diurnal layer, a topic which merits further investigation.
Future work will focus on reproducing the observed diurnal
variability using a 1-D turbulence model, which allows for
the selection of different light absorption schemes, amongst
others. This will allow identification of the dominating terms
in the ocean and atmospheric heat fluxes for each of the regions.
Acknowledgements. This study was funded by the ESA STSE
SSTDV: R.EX.–IM.A.M project, supporting Ioanna Karagali.
Jacob Høyer was partly supported by the Sargasso Eel project,
funded by the Carlsberg Foundation. SEVIRI data were processed
by the Centre de Météorolgie Spatiale, Météo France and the
OSI-SAF project.
Edited by: J. M. Huthnance
References
Bellenger, H. and Duvel, J. P.: An analysis of ocean diurnal warm
layers over tropical oceans, J. Climate, 22, 3629–3646, 2009.
Bellenger, H., Takayabu, Y. N., Ushiyama, T., and Yoneyama, K.:
Role of diurnal warm layers in the diurnal cycle of convection
over the tropical Indian Ocean during MISMO, Mon. Weather
Rev., 138, 2426–2433, 2010.
Castro, S., Wick, G., and Buck, J. J. H.: Comparison of diurnal
warming estimates from unpumped ARGO data and SEVIRI
satellite observations, Remote Sens. Environ., 140, 789–799,
2014.
Clayson, C. A. and Bogdanoff, A. S.: The effect of diurnal sea
surface temperature warming on climatological air–sea fluxes, J.
Climate, 26, 2546–2556, 2013.
Donlon, C., Robinson, I., Casey, K. S., Vasquez-Cuervo, J., Armstrong, E., Arino, O., Gentemann, C., May, D., LeBorgne, P., Piollé, J., Barton, I., Beggs, H., Poulter, D. J. S., Merchant, C. J.,
www.ocean-sci.net/10/745/2014/
757
Bingham, A., Heinz, S., Harris, A., Wick, G., Emery, B., Minnett, P., Evans, R., Llewellyn-Jones, D., Mutlow, C., Reynolds,
R. W., Kawamura, H., and Rayner, N.: The global ocean data
assimilation experiment high-resolution sea surface temperature
pilot project, B. Am. Meteorol. Soc., 88, 1197–1213, 2007.
Eastwood, S., Le Borgne, P., Péré, S., and Poulter, D.: Diurnal variability in sea surface temperature in the Arctic, Remote Sens.
Environ., 115, 2594–2602, 2011.
Embury, O., Merchant, C. J., and Corlett, G. K.: A reprocessing for
climate of sea surface temperature from the along-track scanning
radiometers: initial validation, accounting for skin and diurnal
variability effects, Remote Sens. Environ., 116, 62–78, 2012.
Filipiak, M. J., Merchant, C. J., Kettle, H., and Le Borgne, P.: An
empirical model for the statistics of sea surface diurnal warming,
Ocean Sci., 8, 197–209, doi:10.5194/os-8-197-2012, 2012.
Karagali, I. and Høyer, J.: Observations and modelling of the diurnal SST cycle in the North and Baltic Seas, J. Geophys. Res.Oceans 118, 1–16, 2013.
Karagali, I., Hoeyer, J., and Hasager, C. B.: SST diurnal variability
in the North Sea and the Baltic Sea, Remote Sens. Environ., 112,
159–170, 2012.
Kennedy, J. J., Brohan, P., and Tett, S. F. B.: A global climatology of
the diurnal variations in sea-surface temperature and implications
for MSU temperature trends, Geophys. Res. Lett. 34, L05712,
doi:10.1029/2006GL028920, 2007.
Le Borgne, P., Roquet, H., and Merchant, C. J.: Estimation of sea
surface temperature from the spinning enhanced visible and infrared imager improved using numerical weather prediction, Remote Sens. Environ., 115, 55–65, 2011.
Le Borgne, P., Marsouin, A., Orain, F., and Roquet, H.: Operational
Sea Surface Temperature bias adjustment using AATSR data, Remote Sens. Environ., 116, 93–106, 2012a.
Le Borgne, P., Legendre, G., and Péré, S.: Comparison of
MSG/SEVIRI and drifting buoy derived diurnal warming estimates, Remote Sens. Environ., 124, 622–626, 2012b.
Marullo, S., Santoleri, R., Ciani, D., Le Borgne, P., Péré, S., Pinardi,
N., Tonani, M., and Nardone, G.: Combining model and geostationary satellite data to reconstruct hourly SST field over the
Mediterranean Sea, Remote Sens. Environ., 146, 11–23, 2014.
Merchant, C. J., Filipiak, M. J., Le Borgne, P., Roquet, H., Autret,
E., Piollé, J.-F., and Lavender, S.: Diurnal warm-layer events in
the western Mediterranean and European shelf seas, Geophys.
Res. Lett., 35, L04601, doi:10.1029/2007GL033071, 2008.
Morel, A., Huot, Y., Gentili, B., Werdell, P. J., Hooker, S. B., and
Franz, B. A.: Examining the consistency of products derived
from various ocean color sensors in open ocean (Case 1) waters in the perspective of a multi-sensor approach, Remote Sens.
Environ., 111, 69–88, 2014.
O’Carroll, A. G., Eyre, J. R., and Saunders, R. W.: Three-way error
analysis between AATSR, AMSR-E and in situ Sea Surface Temperature observations, J. Atmos. Ocean. Tech., 25, 1197–1207,
2008.
Ocean & Sea Ice SAF: Atlantic Sea Surface Temperature Product
Manual, Version 1.6, October 2006.
Price, J., Weller, R., Bowers, C., and Briscoe, M. G.: Diurnal response of sea surface temperature observed at the long-term upper ocean study (34◦ N, 70◦ W) in the Sargasso Sea, J. Geophys.
Res., 92, 14480–14490, 1987.
Ocean Sci., 10, 745–758, 2014
758
I. Karagali and J. L. Høyer: SEVIRI regional diurnal warming
Prytherch, J., Farrar, J. T., and Weller, R. A.: Moored surface buoy
observations of the diurnal warm layer, J. Geophys. Res., 118,
4553–4569, 2013.
Richter, I. and Mechoso, C. R.: Orographic influences on the annual
cycle of Namibian stratocumulus clouds, Geophys. Res. Lett.,
31, 1–4, doi:10.1029/2004GL020814, 2004.
Ocean Sci., 10, 745–758, 2014
Ward, B.: Near-surface ocean temperature, J. Geophys. Res. 111,
C02004, doi:10.1029/2004JC002689, 2006.
Webster, P. J., Clayson, C. A., and Curry, J. A.: Clouds, radiation,
and the diurnal cycle of sea surface temperature in the Tropical
Western Pacific, J. Climate, 9, 1712–1730, 1996.
www.ocean-sci.net/10/745/2014/